v0.5 — Parallel Sandboxes, LLM Skills, AI-written PRs, HITL Escalation

Your AI team,
one command away

Describe your project. Pick a theme. Toggle YOLO mode.
Agents work in parallel sandboxes — each on its own branch, building, testing, and shipping PRs autonomously.

Get Started View on GitHub
$ npm i -g @matanelcohen/openspace-ai && openspace 📋 Copied!
npm version
openspace.ai — Diablo Squad
🔥 Diablo Squad
⚔️ Tyrael
🧙 Deckard Cain
🏹 Valla
🛡️ Johanna
Workspaces
📁 e-commerce api
📁 mobile app
👤
You
Set up the auth module with JWT and refresh tokens
⚔️
Tyrael
The forces of order shall prevail. I'll forge the AuthService with bcrypt hashing and token rotation. Expect a PR within minutes.
🧙
Deckard Cain
Stay a while and listen… I've reviewed Tyrael's schema. The refresh_tokens table needs an index on expires_at. I'll add it.
🏹
Valla
Tests locked and loaded. 14 auth specs passing, coverage at 94%. npm test -- auth ✓
Active Tasks
Auth service + JWT
In Progress ⚔️ Tyrael
DB schema review
Done 🧙 Deckard Cain
Auth test suite
Done 🏹 Valla
Rate limiting middleware
Queued 🛡️ Johanna
See it in action

The full experience

Dashboard, tasks, chat, team management, traces, and knowledge — all in one place.

localhost:3000
Dashboard Tasks Chat Team Skills Costs Traces Knowledge GitHub Decisions Memories Settings
How it works

Three steps to your AI squad

From zero to a fully themed, code-shipping AI team in under a minute.

📝

Describe

Tell openspace about your project. The LLM analyzes your codebase, tech stack, and goals to understand what needs to be built.

🎭

Cast

Pick a theme — Diablo, Marvel, Ocean's Eleven, or your own. Each agent gets a unique personality, charter, and skill set.

🚀

Ship

Create tasks or toggle YOLO. Lead triages & delegates. Agents work in parallel sandboxes, each on its own branch. Build, test, PR — all automated.

Features

Everything you need to manage AI teams

Real-time collaboration, full observability, and agents that actually ship code.

🚀

YOLO Mode NEW

Toggle autonomous mode — the lead agent scans the board, triages tasks, assigns to specialists, and ships. You just watch.

🌳

Parallel sandboxes NEW

Each task runs in its own git worktree — isolated branch, parallel execution, auto-commit & PR. Delegated tasks get a feature branch with subtask PRs merging into it.

🧠

Smart skill matching NEW

LLM picks the right skills for each task + agent combo. No more random skill selection — Fry gets React skills, Bender gets API skills.

📝

AI-written PRs NEW

When an agent finishes, it auto-creates a PR with an LLM-generated description summarizing the changes, files modified, and approach taken.

🤖

Agentic execution

Agents don't just generate text — they read files, write code, run tests, fix errors, and iterate. Full Copilot SDK sessions with tools.

💬

Real-time team chat

Chat with your AI agents in real-time. Type "@tyrael fix the auth bug" and it auto-creates a task and assigns it.

🎯

Live task progress

Assign tasks and watch agents work in real-time: thinking, reading files, running commands, all streamed to the UI. Full untruncated logs.

🔎

Prompt inspector

See exactly what was sent to the LLM — system prompt, skills, memories, user message. Full transparency. One-click copy.

👫

Lead delegation

Create a task → lead auto-triages → breaks complex work into subtasks → assigns to specialists → feature branch → consolidated PR to main.

🚨

HITL escalation

Agents escalate when unsure — budget limits, destructive operations, ambiguous requirements. Claim, review, approve or reject with audit trail.

💰

Cost tracking

See exactly what you're spending — cost per agent, per task, per model. Daily trends and breakdowns on a dedicated dashboard.

🐙

GitHub integration

Sync issues to tasks, agents create branches and open PRs. Auto-push branches before PR creation. One-click from task to shipped code.

🧠

Agent memory

Agents learn from past work. Memories retrieved via vector similarity and injected into context for relevant tasks.

🔍

Knowledge base + RAG

7 connectors index docs, tasks, commits, PRs, chat, memories, charters. Local embeddings for semantic search.

🌍

Any stack, any language

Auto-detects your project type — Node, .NET, Python, Go, Rust, Java, Ruby, PHP. Agents install deps, build, and test in your ecosystem.

Pick your theme

Your agents, your universe

Every theme creates unique agents with distinct personalities, charters, and coding styles. Or create your own.

🔥 Diablo
Angels, scholars & demon hunters writing code
⚔️
Tyrael
Lead architect · "Justice demands structure"
backend
🧙
Deckard Cain
Code reviewer · "Stay a while and listen"
review
🏹
Valla
Test specialist · "No bug escapes my sight"
testing
🦸 Marvel
Earth's mightiest developers assemble
🛡️
Captain America
Team lead · "I can do this all day"
lead
🤖
Iron Man
Full-stack engineer · "I am... shipping"
fullstack
🧬
Hulk
Perf engineer · "Hulk OPTIMIZE"
perf
🎰 Ocean's Eleven
The smoothest heist crew in software
🎩
Danny Ocean
Strategist · "I have a plan"
planning
💻
Livingston
Systems hacker · "I'm in"
infra
🃏
Rusty
Frontend ace · "Make it look effortless"
frontend
Observability

See everything your agents do

Every LLM call, skill invocation, and memory access — traced and searchable.

📊 Trace Explorer
🧠 LLM: gpt-4o — plan auth module 1.2s
🔧 Skill: create-file → auth.service.ts 0.3s
💾 Memory: store auth architecture decision 0.1s
🧠 LLM: gpt-4o — generate JWT utils 0.9s
🔧 Skill: run-tests → 14 passed 2.1s
🔎 Knowledge Base — Semantic Search
"how does authentication work?"
0.94 JWT auth flow with refresh token rotation auth.service.ts
0.89 Bcrypt password hashing with salt rounds=12 user.model.ts
0.82 Rate limiting middleware for auth endpoints middleware/rate-limit.ts
0.78 Session architecture decision (2024-01-15) memory
Built with

Standing on the shoulders of giants

Powered by best-in-class tools and frameworks.

🤖 Squad SDK ✨ GitHub Copilot ▲ Next.js ⚡ Fastify 🗄️ SQLite

Ready to meet your squad?

One command. Themed AI agents. Real code shipped.

npm i -g @matanelcohen/openspace-ai Star on GitHub ⭐